You mean http://www.wurmonline.com/ ? (A MMO that by no accident notch was working on before Minecraft)

Sorry could not resist :)

To begin their department have a diagnostic space with several axes onto where the classic autism and aspberger diagnoses are just specified areas in their diagnostic space. This is because the classic diagnoses are relatively static and might not pinpoint the correct treatments/remedies.

A big result of this is that they have realized that a lot of women with problems coping in society(crime,etc) in reality fits into parts of that diagnostic spectrum that would have made them fall out of the classic autism/aspergers diagnoses but still retain a multitude of "problems" in interactions with other people.

Let your kid roam on the computer and once it slows down teach him to reinstall the computer himself.

Anti-virus programs are reactive rather than proactive so you should expect a windows machine to be infected soon or later (unless used by a somewhat obsessive noscript,etc user that avoids most risks).

As you mention Teslas i guess using openCL with AMD could be an option?

Since the fusion chips share memory (for better and worse) with the main CPU you can apperantly get faster(0?) "transfer time" between CPU/GPU also maybe with this method it'd be possible to pump up the GPU cores with larger amounts of useable memory than usual?

And since they're cheap you could buy a bucketload of them.

Install your favourite linux on virtualbox and then add the host additions, setup the mapping in VBox and mount the path in linux and presto.. filesystem sharing.

Sure this is a big security issue for networked apps, especially those that listens on sockets.

However for many desktop utility apps (graphics programs,etc) that has a tendency to depend on cutting edge libraries, it's a major hurdle for users to try them out (without borking up half their system to go cutting edge for that single app). This might even be a reason for why linux has never really succeeded on the desktop since you either get a flawed experience or an outdated experience.

As for security, a 0install system like this could black-list insecure library versions until they get upgraded by the authors.

Actually i think you can enable PAE with a bit of hacking.

There are however a few big problems with PAE.

1: Pages in memory are 4mb instead of 4kb, some programs make silly assumptions about it and that decreases compability.
2: Ever more severe, many thirdparty drivers does the same. Thus PAE mode would induce a whole world of hurt in terms of compability and system crashes.
3: Bloat.. since most programs has code, constant, stack and data areas that usually end up on separate pages every minor app will require something like 16meg of memory. Not a big problem for a server with lots of mem and few programs but worse in desktop settings.

So.. when you go to PAE you could as well just jump to 64bit because of the driver and app issues and that will also be more efficient since it's really only the kernel that will consume more memory due to larger pointers.

The apps are still 32bit for most parts but can be put into separate areas with virtual mappings on a small (4k) granularity that doesn't induce bloat.

It's solid for normal serving of servlet stuff but that's about it sadly.

- Your "security record" comes from the fact that tomcat is written in a "safe" language, a security hole would have to come from some stupid manual hole, a JVM bug or the APR connector.

- The non-blocking extensions are non-standard, and much worse than that is that those extensions only works with the APR connector (IE NATIVE C CODE!) but are only experimental with the portable NIO connector (after alot of debugging my own code i finally found a blog post about NIO not working with tomcat async). So it means that you need to be compiling/installing the tomcat code for the target platform before deployment instead of just copying over the JAR's or something like that as you'd be expecting with a Java app.

- The internals are quite contrived, oh i'm sure there's a good reason and prolly has to do with the APR/NIO/"classic" multi-connector support but it made me drop the option of actually fixing the NIO support. (so the open source advantage goes out the door)

- Last i tried (autnum) the API for servlet 3 async was there in tomcat7 but it seems the implementation wasn't? gave me alot of headache and the APR only support for their propeitary async api finally broke most of my faith in tomcat as a future platform.

The most ironic part about all of this is that i decided to do a quick project with a "established" base system and not be hacking things up, but i'd say i rediscovered a truth. If you want to do something new with an established package you might just run into experimental features and/or features that would requie the entire system to be rearchitectured (both in the case of tomcat).

I've now taken a big leap to an "experimental" platform(Node.JS) and while not perfect at all it was built from day one for the kind of things i wanted to do. Oh and working with Javascript on a small project is an absolute pleasure compared to Java :)

No, there is one way to represent every number.

The mantissa doesn't state a "number", it is the decimal part of 1.xxxxx.

Say we have a 2 bit mantissa. The mantissa would thus describe the following numbers:
00 -> 1.00
01 -> 1.25
10 -> 1.5
11 -> 1.75

When we reach 2 we just increase the exponent instead, This gives the double representation a bigger range.

For very small numbers there is also a "subnormal" representation in the IEEE standard. This is when the exponent reaches the "smallest" possible value. At this point the mantissa bits represents numbers from "2" to "0" with the smallest exponent.

If i'm not off my hat right now this is why setting all the bits of a float or double will actually produce a 0, lowest exponent, -> subnormals = actual 0 and no sign(irrelevant in the case of floating point numbers though)

Could you give a reference on this? Partial register stalls on al->ax->eax writing is kinda obvious but this would only apply if you read after a partial write, doing a full register write would remove any partial stalls.

In an analogy to that the partial writing of a the flags shouldn't matter unless you depend on it from an immediate flag reading instruction like a conditional branch or treating flags like numbers (ie pushf,etc). Especially since a shift would likely to be followed by a compare the shifts flag state is irrelevant for the one who reads the flags as it has been overwritten.

It's somewhat ironic that the ARM chipset and architecture was first created by a computer manufacturer that flopped, only to see the cpu architecture that was the offspring actually be the only architecture to ever gain at the expense of the X86 architecture that probably was the culprit of destroying acorn computers.

And with this happening and even more focus being shifted onto the web as being the platform independent delivery platform, it won't be Intel flirting with Apple and Linux that got MS onto the fast path to obscurity. But the availability of a ever so cheap platform, powered by a ghost from the past.